voculinux/tests/test_recognition_manager_downloads.py

"""
Coverage boost tests targeting major gaps in recognition_manager and ibus_engine.

Key focus areas:
- Model download methods with progress tracking
- Audio reconnection logic
- IBus engine utility functions
"""

import base64
import os
import sys
import time
from unittest.mock import MagicMock, patch

import pytest

# Mock GI imports before importing any vocalinux modules that use gi.
# On CI, real gi/IBus packages are installed; without mocks, importing
# ibus_engine would connect to a real IBus daemon and hang.
if "gi" not in sys.modules:
    sys.modules["gi"] = MagicMock()
if "gi.repository" not in sys.modules:
    sys.modules["gi.repository"] = MagicMock()

from vocalinux.common_types import RecognitionState
from vocalinux.speech_recognition.recognition_manager import SpeechRecognitionManager
from vocalinux.utils.faster_whisper_model_info import manifest_key as faster_whisper_manifest_key
from vocalinux.utils.faster_whisper_model_info import model_files as faster_whisper_model_files
from vocalinux.utils.model_checksums import VERIFICATION_STAMP_NAME, ChecksumError, expected_for
from vocalinux.utils.model_checksums import verify_model_file as verify_model_file_real
from vocalinux.utils.parakeet_model_info import MODEL_FILES as PARAKEET_MODEL_FILES
from vocalinux.utils.parakeet_model_info import manifest_key as parakeet_manifest_key


def _make_manager(engine="whisper_cpp", **kw):
    """Create a SpeechRecognitionManager with mocked initialization."""
    with patch.object(SpeechRecognitionManager, "_init_vosk"):
        with patch.object(SpeechRecognitionManager, "_init_whisper"):
            with patch.object(SpeechRecognitionManager, "_init_whispercpp"):
                with patch.object(SpeechRecognitionManager, "_init_parakeet"):
                    with patch.object(SpeechRecognitionManager, "_init_faster_whisper"):
                        mgr = SpeechRecognitionManager(
                            engine=engine,
                            model_size="small",
                            language="en-us",
                            defer_download=True,
                            **kw,
                        )
                    # Ensure vosk_model_map is set (normally done in _init_vosk)
                    if not hasattr(mgr, "vosk_model_map"):
                        # The names _init_vosk() would pick for en-us. They have to be
                        # real: the download path looks each one up in the checksum
                        # manifest to stamp the tree it extracts.
                        mgr.vosk_model_map = {
                            "small": "vosk-model-small-en-us-0.15",
                            "medium": "vosk-model-en-us-0.22",
                            "large": "vosk-model-en-us-0.22",
                        }
                    return mgr


# A real zip, embedded rather than built here: by the time this module is
# imported, other test modules have already replaced zipfile/BytesIO in
# sys.modules with mocks, so building one here yields empty bytes. Its single
# entry sits under the directory VOSK's small en-us archive unpacks to, because
# that is the tree the downloader stamps after extracting.
VOSK_ZIP_BYTES = base64.b64decode(
    "UEsDBBQAAAAIAHV2F12oDuXYFgAAAIgTAAAoAAAAdm9zay1tb2RlbC1zbWFsbC1lbi11cy0wLjE1L2FtL2Zp"
    "bmFsLm1kbO3BMQEAAADCoPVPbQo/oAAAAACAtwFQSwECFAMUAAAACAB1dhddqA7l2BYAAACIEwAAKAAAAAAA"
    "AAAAAAAAgAEAAAAAdm9zay1tb2RlbC1zbWFsbC1lbi11cy0wLjE1L2FtL2ZpbmFsLm1kbFBLBQYAAAAAAQAB"
    "AFYAAABcAAAAAAA="
)


MODEL_NAME = "vosk-model-small-en-us-0.15"

# Same shape, but its top-level directory is not the model name: the downloader
# has to notice rather than report a success that installed nothing.
WRONG_LAYOUT_ZIP_BYTES = base64.b64decode(
    "UEsDBBQAAAAIAHeGF11E9MkCEQAAANAHAAAcAAAAc29tZS1vdGhlci1uYW1lL2FtL2ZpbmFsLm1kbGNgGAWj"
    "YBSMglEwCkbBUAcAUEsBAhQDFAAAAAgAd4YXXUT0yQIRAAAA0AcAABwAAAAAAAAAAAAAAIABAAAAAHNvbWUt"
    "b3RoZXItbmFtZS9hbS9maW5hbC5tZGxQSwUGAAAAAAEAAQBKAAAASwAAAAAA"
)


class FakeRequestError(Exception):
    """Stands in for requests.exceptions.RequestException.

    A leaked mock makes ``except requests.exceptions.RequestException`` a
    TypeError, so the mocked module needs a real class here. It must not be
    ``Exception`` itself, or that clause swallows every other failure the
    downloader is supposed to surface unwrapped.
    """


def _fake_clock(step=0.2):
    """Monotonic stand-in for time.time().

    A fixed side_effect list runs out unpredictably: time.time is patched
    globally, so logging consumes ticks too.
    """
    from itertools import count

    counter = count()
    return lambda: next(counter) * step


@pytest.fixture
def skip_checksum():
    """Accept the synthetic payloads these tests stream.

    Downloads are verified against the digests pinned in model_checksums.txt, so
    a few hundred bytes of ``b"x"`` are correctly rejected. These tests cover
    download *mechanics* (progress, content-length, URL shaping); integrity
    itself is covered by tests/test_model_checksums.py. Yielding the mock lets
    each test still assert that verification was reached.
    """
    with patch("vocalinux.speech_recognition.recognition_manager.verify_model_file") as mock_verify:
        yield mock_verify


@pytest.fixture(autouse=True)
def cleanup_sys_modules():
    """Cleanup sys.modules after each test - full snapshot/restore."""
    # Take a complete snapshot of sys.modules before the test
    saved_modules = dict(sys.modules)

    yield

    # Restore sys.modules to exact pre-test state
    added_keys = set(sys.modules.keys()) - set(saved_modules.keys())
    for key in added_keys:
        del sys.modules[key]
    for key, value in saved_modules.items():
        if key not in sys.modules or sys.modules[key] is not value:
            sys.modules[key] = value


class TestDownloadWhispercppModel:
    """Test _download_whispercpp_model() with runtime import mocking."""

    def test_download_whispercpp_success_basic(self, tmp_path, skip_checksum):
        """Test successful whisper.cpp model download."""
        manager = _make_manager(engine="whisper_cpp")
        model_file = str(tmp_path / "ggml-small.bin")

        mock_requests = MagicMock()
        mock_response = MagicMock()
        mock_response.headers = {"content-length": "1000"}
        mock_response.iter_content.return_value = [b"x" * 500, b"y" * 500]
        mock_requests.get.return_value = mock_response
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.get_model_path",
                return_value=model_file,
            ):
                manager._download_whispercpp_model()

        assert os.path.exists(model_file)
        assert os.path.getsize(model_file) == 1000
        # The model is only installed after it is verified.
        skip_checksum.assert_called_once()

    def test_download_whispercpp_progress_callback(self, tmp_path, skip_checksum):
        """Test progress callback is invoked during download."""
        manager = _make_manager(engine="whisper_cpp")
        progress_calls = []

        def track_progress(progress, speed, status):
            progress_calls.append((progress, speed, status))

        manager._download_progress_callback = track_progress
        model_file = str(tmp_path / "ggml-small.bin")

        mock_requests = MagicMock()
        mock_response = MagicMock()
        mock_response.headers = {"content-length": "1000"}
        mock_response.iter_content.return_value = [b"x" * 500, b"y" * 500]
        mock_requests.get.return_value = mock_response
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.get_model_path",
                return_value=model_file,
            ):
                with patch("time.time", side_effect=_fake_clock()):
                    manager._download_whispercpp_model()

        mock_requests.get.assert_called_once()
        call_args = mock_requests.get.call_args
        assert call_args is not None
        assert len(call_args[0]) > 0 or "url" in call_args[1]
        assert len(progress_calls) >= 1

    def test_download_whispercpp_no_content_length(self, tmp_path, skip_checksum):
        """Test download when content-length header is missing."""
        manager = _make_manager(engine="whisper_cpp")
        model_file = str(tmp_path / "ggml-small.bin")

        mock_requests = MagicMock()
        mock_response = MagicMock()
        mock_response.headers = {}  # No content-length
        mock_response.iter_content.return_value = [b"x" * 500, b"y" * 500]
        mock_requests.get.return_value = mock_response
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.get_model_path",
                return_value=model_file,
            ):
                manager._download_whispercpp_model()

        assert os.path.exists(model_file)

    def test_download_whispercpp_request_error(self, tmp_path):
        """Test download request error handling."""
        manager = _make_manager(engine="whisper_cpp")
        model_file = str(tmp_path / "ggml-small.bin")

        mock_requests = MagicMock()
        mock_error = Exception("Network error")
        mock_requests.get.side_effect = mock_error
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.get_model_path",
                return_value=model_file,
            ):
                with pytest.raises(RuntimeError, match="Failed to download"):
                    manager._download_whispercpp_model()

    def test_download_whispercpp_appends_download_true(self, tmp_path, skip_checksum):
        """Hugging Face URLs get ?download=true for reliable binary responses."""
        manager = _make_manager(engine="whisper_cpp")
        model_file = str(tmp_path / "ggml-small.bin")

        mock_requests = MagicMock()
        mock_response = MagicMock()
        mock_response.headers = {"content-length": "4", "content-type": "application/octet-stream"}
        mock_response.iter_content.return_value = [b"data"]
        mock_requests.get.return_value = mock_response
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.get_model_path",
                return_value=model_file,
            ):
                manager._download_whispercpp_model()

        called_url = mock_requests.get.call_args[0][0]
        assert "huggingface.co" in called_url
        assert "download=true" in called_url
        assert mock_requests.get.call_args[1].get("timeout") == manager._MODEL_DOWNLOAD_TIMEOUT

    def test_download_whispercpp_timeout_message(self, tmp_path):
        """Timeout-like errors surface a dedicated user-facing message."""
        manager = _make_manager(engine="whisper_cpp")
        model_file = str(tmp_path / "ggml-small.bin")

        mock_requests = MagicMock()
        mock_requests.get.side_effect = Exception("Read timeout")
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.get_model_path",
                return_value=model_file,
            ):
                with pytest.raises(RuntimeError, match="timed out"):
                    manager._download_whispercpp_model()

    def test_stream_model_download_rejects_html(self, tmp_path):
        """HTML error pages must not be written as model binaries."""
        manager = _make_manager(engine="whisper_cpp")
        dest = str(tmp_path / "bad.bin")

        mock_requests = MagicMock()
        mock_response = MagicMock()
        mock_response.headers = {"content-type": "text/html; charset=utf-8"}
        mock_response.status_code = 200
        mock_requests.get.return_value = mock_response

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with pytest.raises(RuntimeError, match="HTML"):
                manager._stream_model_download("https://example.com/model.bin", dest)

        assert not os.path.exists(dest)

    def test_stream_model_download_empty_body(self, tmp_path):
        """Zero-byte downloads are treated as failure and cleaned up."""
        manager = _make_manager(engine="whisper_cpp")
        dest = str(tmp_path / "empty.bin")

        mock_requests = MagicMock()
        mock_response = MagicMock()
        mock_response.headers = {"content-length": "0", "content-type": "application/octet-stream"}
        mock_response.iter_content.return_value = [b"", b""]
        mock_requests.get.return_value = mock_response

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with pytest.raises(RuntimeError, match="0 bytes"):
                manager._stream_model_download("https://example.com/model.bin", dest)

        assert not os.path.exists(dest)

    def test_stream_model_download_cancelled(self, tmp_path):
        """User cancel mid-stream removes the partial file."""
        manager = _make_manager(engine="whisper_cpp")
        manager._download_cancelled = True
        dest = str(tmp_path / "partial.bin")

        mock_requests = MagicMock()
        mock_response = MagicMock()
        mock_response.headers = {
            "content-length": "100",
            "content-type": "application/octet-stream",
        }
        mock_response.iter_content.return_value = [b"chunk"]
        mock_requests.get.return_value = mock_response

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with pytest.raises(RuntimeError, match="cancelled"):
                manager._stream_model_download("https://example.com/model.bin", dest)

        assert not os.path.exists(dest)

    def test_stream_model_download_eta_minutes_and_empty_chunks(self, tmp_path):
        """Progress path covers multi-minute ETA and skips empty chunks."""
        manager = _make_manager(engine="whisper_cpp")
        progress_calls = []
        manager._download_progress_callback = lambda progress, speed, status: progress_calls.append(
            status
        )
        dest = str(tmp_path / "big.bin")

        mock_requests = MagicMock()
        mock_response = MagicMock()
        # Large total so remaining/speed yields ETA >= 60s
        mock_response.headers = {
            "content-length": str(100 * 1024 * 1024),
            "content-type": "application/octet-stream",
        }
        mock_response.iter_content.return_value = [b"", b"x" * 1024]
        mock_requests.get.return_value = mock_response

        # start=0, then 0.2 for progress update (elapsed > 0, small speed)
        times = [0.0, 0.2, 0.2]
        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch("time.time", side_effect=lambda: times.pop(0) if times else 1.0):
                manager._stream_model_download("https://example.com/model.bin", dest)

        assert os.path.exists(dest)
        assert os.path.getsize(dest) == 1024
        assert any("m " in s or "ETA" in s for s in progress_calls)


class TestDownloadWhisperModel:
    """OpenAI Whisper downloads run through the shared streaming helper.

    Vocalinux fetches the checkpoint only to give the UI progress and a working
    cancel; whisper.load_model() is what verifies it. Both were briefly lost when
    the downloader was removed altogether, so they are pinned here.
    """

    @staticmethod
    def _fake_stream(url, dest_path):
        with open(dest_path, "wb") as handle:
            handle.write(b"checkpoint")

    def test_writes_the_name_whisper_expects(self, tmp_path):
        manager = _make_manager(engine="whisper")
        manager.model_size = "large"

        with patch.object(manager, "_stream_model_download", side_effect=self._fake_stream):
            manager._download_whisper_model(str(tmp_path))

        # large is stored as large-v3.pt; "large.pt" here would mean load_model
        # downloads the 2.9GB checkpoint a second time.
        assert (tmp_path / "large-v3.pt").exists()
        assert not (tmp_path / "large.pt").exists()

    def test_reports_progress(self, tmp_path):
        manager = _make_manager(engine="whisper")
        progress_calls = []
        manager._download_progress_callback = lambda f, s, st: progress_calls.append(st)

        with patch.object(manager, "_stream_model_download", side_effect=self._fake_stream):
            manager._download_whisper_model(str(tmp_path))

        assert progress_calls, "the download dialog would sit at zero"

    def test_cancel_propagates_and_cleans_up(self, tmp_path):
        """Cancel raised by the stream helper must reach the caller."""
        manager = _make_manager(engine="whisper")

        def cancel(url, dest_path):
            with open(dest_path, "wb") as handle:
                handle.write(b"partial")
            raise RuntimeError("Download cancelled")

        # A leaked requests mock from another module makes
        # `except requests.exceptions.RequestException` a TypeError, so pin a
        # real exception class here as the other download tests do.
        mock_requests = MagicMock()
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch.object(manager, "_stream_model_download", side_effect=cancel):
                with pytest.raises(RuntimeError, match="cancelled"):
                    manager._download_whisper_model(str(tmp_path))

        assert not list(tmp_path.glob("*"))


class TestDownloadVoskModel:
    """Test _download_vosk_model() with runtime import mocking."""

    @staticmethod
    def _serve_zip(manager, tmp_path, payload=None):
        """Run a full download+extract of an archive into tmp_path."""
        payload = VOSK_ZIP_BYTES if payload is None else payload
        mock_requests = MagicMock()
        mock_response = MagicMock()
        mock_response.headers = {"content-length": str(len(payload))}
        mock_response.iter_content.return_value = [payload]
        mock_requests.get.return_value = mock_response
        mock_requests.exceptions.RequestException = FakeRequestError

        # Extraction has to really happen here — the stamp lands in the tree it
        # creates — and other modules leave a MagicMock in sys.modules["zipfile"].
        real_zipfile = getattr(sys, "_vocalinux_real_zipfile", None) or __import__("zipfile")

        with patch.dict("sys.modules", {"requests": mock_requests, "zipfile": real_zipfile}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.MODELS_DIR", str(tmp_path)
            ):
                with patch("time.time", side_effect=_fake_clock()):
                    manager._download_vosk_model()
        return mock_requests

    def test_download_vosk_progress_callback(self, tmp_path, skip_checksum):
        """Test progress callback during Vosk download."""
        manager = _make_manager(engine="vosk")
        progress_calls = []

        def track_progress(progress, speed, status):
            progress_calls.append((progress, speed, status))

        manager._download_progress_callback = track_progress

        mock_requests = self._serve_zip(manager, tmp_path)

        mock_requests.get.assert_called_once()
        call_args = mock_requests.get.call_args
        assert call_args is not None
        assert len(call_args[0]) > 0 or "url" in call_args[1]
        assert len(progress_calls) >= 1
        # Otherwise the digest check could be deleted and this stay green.
        skip_checksum.assert_called_once()

    def test_download_vosk_stamps_the_extracted_tree(self, tmp_path, skip_checksum):
        """A directory has no digest; install.sh reads this stamp instead.

        Leave it out and every model fetched at first run or from Settings looks
        unverified to the next ./install.sh, which downloads it again.
        """
        manager = _make_manager(engine="vosk")

        self._serve_zip(manager, tmp_path)

        stamp = tmp_path / "vosk-model-small-en-us-0.15" / VERIFICATION_STAMP_NAME
        pinned = expected_for("vosk-model-small-en-us-0.15.zip")
        assert pinned is not None, "the fixture model must be in the manifest"
        assert stamp.read_text().strip() == pinned.digest

    @staticmethod
    def _leftovers(tmp_path):
        """Everything in MODELS_DIR that is not the finished model tree."""
        return sorted(p.name for p in tmp_path.iterdir() if p.name != MODEL_NAME)

    def test_download_vosk_fails_when_the_stamp_cannot_be_written(self, tmp_path, skip_checksum):
        """An unstampable tree is one we would silently refetch; say so instead."""
        manager = _make_manager(engine="vosk")

        with patch(
            "vocalinux.speech_recognition.recognition_manager.write_verification_stamp",
            side_effect=OSError("read-only models directory"),
        ):
            with pytest.raises(OSError):
                self._serve_zip(manager, tmp_path)

        # And it must not leave the tree behind: _init_vosk would call that
        # installed while install.sh, finding no stamp, refetched it every run.
        assert not (tmp_path / MODEL_NAME).exists()
        assert self._leftovers(tmp_path) == []

    def test_a_failed_unpack_keeps_the_model_that_was_already_there(self, tmp_path, skip_checksum):
        """Replacing is a swap: a broken download must not cost the old model."""
        old = tmp_path / MODEL_NAME / "am"
        old.mkdir(parents=True)
        (old / "final.mdl").write_bytes(b"the model that was already installed")

        manager = _make_manager(engine="vosk")
        with patch(
            "vocalinux.speech_recognition.recognition_manager.write_verification_stamp",
            side_effect=OSError("read-only models directory"),
        ):
            with pytest.raises(OSError):
                self._serve_zip(manager, tmp_path)

        assert (old / "final.mdl").read_bytes() == b"the model that was already installed"
        assert self._leftovers(tmp_path) == []

    def test_an_archive_with_the_wrong_layout_is_refused(self, tmp_path, skip_checksum):
        """A zip that unpacks somewhere else must not look like a success."""
        manager = _make_manager(engine="vosk")

        with pytest.raises(RuntimeError, match="does not contain"):
            self._serve_zip(manager, tmp_path, payload=WRONG_LAYOUT_ZIP_BYTES)

        assert not (tmp_path / MODEL_NAME).exists()
        assert self._leftovers(tmp_path) == []

    def test_a_verified_download_replaces_an_older_tree(self, tmp_path, skip_checksum):
        old = tmp_path / MODEL_NAME
        (old / "am").mkdir(parents=True)
        (old / "am" / "final.mdl").write_bytes(b"stale")
        (old / "stale-file").write_text("must not survive the swap")

        manager = _make_manager(engine="vosk")
        self._serve_zip(manager, tmp_path)

        assert (old / "am" / "final.mdl").read_bytes() != b"stale"
        assert not (old / "stale-file").exists(), "the swap must replace, not merge"
        assert (old / VERIFICATION_STAMP_NAME).exists()
        assert self._leftovers(tmp_path) == []

    def test_download_vosk_request_error(self, tmp_path):
        """Test Vosk download request error handling."""
        manager = _make_manager(engine="vosk")

        mock_requests = MagicMock()
        mock_error = Exception("Network error")
        mock_requests.get.side_effect = mock_error
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.MODELS_DIR", str(tmp_path)
            ):
                with pytest.raises(RuntimeError, match="Failed to download"):
                    manager._download_vosk_model()


class TestWhispercppRejectsAnUnverifiedModelOnDisk:
    """Wiring test: the hash has to happen where the model is picked up.

    _download_whispercpp_model verifies before its rename, so the exposure is a
    file that never came through it: releases before verification fetched ggml
    models from `resolve/main` unchecked, and install.sh re-hashes only
    ggml-tiny.bin. Without this the file goes to pywhispercpp/ctypes as-is.
    """

    def test_a_model_that_fails_its_pin_is_removed_and_not_loaded(self, tmp_path):
        model = tmp_path / "ggml-tiny.bin"
        model.write_bytes(b"not the model that is pinned")

        manager = _make_manager(engine="whisper_cpp")
        manager._defer_download = True

        mock_pywhispercpp = MagicMock()
        with patch.dict(
            "sys.modules",
            {"pywhispercpp": mock_pywhispercpp, "pywhispercpp.model": mock_pywhispercpp},
        ):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.get_model_path",
                return_value=str(model),
            ):
                with patch.object(manager, "_load_whispercpp_model") as load:
                    manager._init_whispercpp()

        assert not model.exists(), "an unverifiable model must not stay on disk"
        load.assert_not_called()
        assert manager._model_initialized is False

    def test_a_model_that_cannot_be_deleted_is_still_not_loaded(self, tmp_path):
        model = tmp_path / "ggml-tiny.bin"
        model.write_bytes(b"not the model that is pinned")

        manager = _make_manager(engine="whisper_cpp")
        manager._defer_download = False

        mock_pywhispercpp = MagicMock()
        with patch.dict(
            "sys.modules",
            {"pywhispercpp": mock_pywhispercpp, "pywhispercpp.model": mock_pywhispercpp},
        ):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.get_model_path",
                return_value=str(model),
            ):
                with patch(
                    "vocalinux.speech_recognition.recognition_manager.os.remove",
                    side_effect=OSError("read-only"),
                ):
                    with patch.object(manager, "_load_whispercpp_model") as load:
                        with pytest.raises(RuntimeError, match="failed verification"):
                            manager._init_whispercpp()

        assert model.exists()
        load.assert_not_called()

    def test_an_unpinned_model_is_refused_not_deleted(self, tmp_path):
        model = tmp_path / "ggml-not-in-manifest.bin"
        model.write_bytes(b"bytes")

        manager = _make_manager(engine="whisper_cpp")
        manager._defer_download = True

        mock_pywhispercpp = MagicMock()
        with patch.dict(
            "sys.modules",
            {"pywhispercpp": mock_pywhispercpp, "pywhispercpp.model": mock_pywhispercpp},
        ):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.get_model_path",
                return_value=str(model),
            ):
                with patch.object(manager, "_load_whispercpp_model") as load:
                    manager._init_whispercpp()

        assert model.exists(), "a missing pin is not a reason to delete the file"
        load.assert_not_called()
        assert manager._model_initialized is False


class TestParakeetRejectsAnUnverifiedModelOnDisk:
    """Wiring test: the hash has to happen where the model is picked up.

    _download_parakeet_model verifies before its rename, so the exposure is a
    bundle that never came through it: files copied in, or left over from a
    cancelled download. sherpa-onnx loads these through native code. Without
    this the files go to OfflineRecognizer.from_transducer as-is.
    """

    @staticmethod
    def _write_bundle(tmp_path, payload=b"not the model that is pinned"):
        model_dir = tmp_path / "v3-european"
        model_dir.mkdir()
        for name in PARAKEET_MODEL_FILES:
            (model_dir / name).write_bytes(payload)
        return model_dir

    def test_a_model_that_fails_its_pin_is_removed_and_not_loaded(self, tmp_path):
        model_dir = self._write_bundle(tmp_path)

        manager = _make_manager(engine="parakeet")
        manager.model_size = "v3-european"
        manager._defer_download = True

        mock_sherpa = MagicMock()
        with patch.dict("sys.modules", {"sherpa_onnx": mock_sherpa}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.parakeet.get_model_path",
                return_value=str(model_dir),
            ):
                manager._init_parakeet()

        for name in PARAKEET_MODEL_FILES:
            assert not (model_dir / name).exists(), "an unverifiable model must not stay on disk"
        mock_sherpa.OfflineRecognizer.from_transducer.assert_not_called()
        assert manager._model_initialized is False

    def test_a_model_that_cannot_be_deleted_is_still_not_loaded(self, tmp_path):
        model_dir = self._write_bundle(tmp_path)

        manager = _make_manager(engine="parakeet")
        manager.model_size = "v3-european"
        manager._defer_download = False

        mock_sherpa = MagicMock()
        with patch.dict("sys.modules", {"sherpa_onnx": mock_sherpa}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.parakeet.get_model_path",
                return_value=str(model_dir),
            ):
                with patch(
                    "vocalinux.speech_recognition.recognition_manager.os.remove",
                    side_effect=OSError("read-only"),
                ):
                    with pytest.raises(RuntimeError, match="failed verification"):
                        manager._init_parakeet()

        for name in PARAKEET_MODEL_FILES:
            assert (model_dir / name).exists()
        mock_sherpa.OfflineRecognizer.from_transducer.assert_not_called()

    def test_an_unpinned_model_is_refused_not_deleted(self, tmp_path):
        """Parakeet pins are constructed keys; simulate a missing pin for one file."""
        model_dir = self._write_bundle(tmp_path, payload=b"bytes")
        unpinned_name = PARAKEET_MODEL_FILES[0]
        unpinned_key = parakeet_manifest_key("v3-european", unpinned_name)

        def fake_verify(path, filename=None):
            key = filename or os.path.basename(path)
            if key == unpinned_key:
                raise ChecksumError(f"No checksum is pinned for {key}")
            verify_model_file_real(path, filename)

        def fake_expected(filename):
            if os.path.basename(filename) == unpinned_key:
                return None
            return expected_for(filename)

        manager = _make_manager(engine="parakeet")
        manager.model_size = "v3-european"
        manager._defer_download = True

        mock_sherpa = MagicMock()
        with patch.dict("sys.modules", {"sherpa_onnx": mock_sherpa}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.parakeet.get_model_path",
                return_value=str(model_dir),
            ):
                with patch(
                    "vocalinux.speech_recognition.recognition_manager.verify_model_file",
                    side_effect=fake_verify,
                ):
                    with patch(
                        "vocalinux.speech_recognition.recognition_manager.expected_for",
                        side_effect=fake_expected,
                    ):
                        manager._init_parakeet()

        assert (
            model_dir / unpinned_name
        ).exists(), "a missing pin is not a reason to delete the file"
        mock_sherpa.OfflineRecognizer.from_transducer.assert_not_called()
        assert manager._model_initialized is False


class TestParakeetDownloadVerifiesExistingFiles:
    """Existence is not a pin: a leftover dest must still match its digest.

    A previous continue-on-exists skipped verification and left a copied-in or
    truncated file for sherpa-onnx to load. OfflineRecognizer is not involved
    here; this is the download loop only.
    """

    def test_an_existing_file_that_fails_its_pin_is_redownloaded(self, tmp_path):
        manager = _make_manager(engine="parakeet")
        manager.model_size = "v3-european"
        model_dir = tmp_path / "v3-european"
        model_dir.mkdir()

        encoder_name = PARAKEET_MODEL_FILES[0]
        encoder = model_dir / encoder_name
        encoder.write_bytes(b"not the encoder that is pinned")
        encoder_key = parakeet_manifest_key("v3-european", encoder_name)

        streamed = []
        verify_calls = []

        def fake_stream(url, dest_path):
            streamed.append(dest_path)
            with open(dest_path, "wb") as handle:
                handle.write(b"good-enough")

        def fake_verify(path, filename=None):
            verify_calls.append((path, filename))
            # Existing dests fail; freshly streamed temps pass.
            if not str(path).endswith(".tmp"):
                raise ChecksumError("digest mismatch")

        mock_requests = MagicMock()
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.parakeet.get_model_path",
                return_value=str(model_dir),
            ):
                with patch.object(manager, "_stream_model_download", side_effect=fake_stream):
                    with patch(
                        "vocalinux.speech_recognition.recognition_manager.verify_model_file",
                        side_effect=fake_verify,
                    ):
                        manager._download_parakeet_model()

        assert any(
            path == str(encoder) and filename == encoder_key for path, filename in verify_calls
        ), "an existing dest must be hashed, not skipped because it is already on disk"
        assert any(
            os.path.basename(path) == encoder_name + ".tmp" for path in streamed
        ), "an existing bad file must be removed and re-downloaded"
        assert encoder.read_bytes() == b"good-enough"


class TestFasterWhisperRejectsAnUnverifiedModelOnDisk:
    """Wiring test: the hash has to happen where the model is picked up."""

    @staticmethod
    def _write_bundle(tmp_path, payload=b"not the model that is pinned"):
        model_dir = tmp_path / "tiny"
        model_dir.mkdir()
        for name in faster_whisper_model_files("tiny"):
            (model_dir / name).write_bytes(payload)
        return model_dir

    def test_a_model_that_fails_its_pin_is_removed_and_not_loaded(self, tmp_path):
        model_dir = self._write_bundle(tmp_path)

        manager = _make_manager(engine="faster_whisper")
        manager.model_size = "tiny"
        manager._defer_download = True

        with patch(
            "vocalinux.speech_recognition.recognition_manager.faster_whisper.get_model_path",
            return_value=str(model_dir),
        ):
            with patch(
                "vocalinux.speech_recognition.engines.faster_whisper_engine.FasterWhisperEngine"
            ) as engine_cls:
                manager._init_faster_whisper()

        for name in faster_whisper_model_files("tiny"):
            assert not (model_dir / name).exists(), "an unverifiable model must not stay on disk"
        engine_cls.assert_not_called()
        assert manager._model_initialized is False

    def test_a_model_that_cannot_be_deleted_is_still_not_loaded(self, tmp_path):
        model_dir = self._write_bundle(tmp_path)

        manager = _make_manager(engine="faster_whisper")
        manager.model_size = "tiny"
        manager._defer_download = False

        with patch(
            "vocalinux.speech_recognition.recognition_manager.faster_whisper.get_model_path",
            return_value=str(model_dir),
        ):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.os.remove",
                side_effect=OSError("read-only"),
            ):
                with patch(
                    "vocalinux.speech_recognition.engines.faster_whisper_engine.FasterWhisperEngine"
                ) as engine_cls:
                    with pytest.raises(RuntimeError, match="failed verification"):
                        manager._init_faster_whisper()

        for name in faster_whisper_model_files("tiny"):
            assert (model_dir / name).exists()
        engine_cls.assert_not_called()

    def test_an_unpinned_model_is_refused_not_deleted(self, tmp_path):
        """faster-whisper pins are constructed keys; simulate a missing pin for one file."""
        model_dir = self._write_bundle(tmp_path, payload=b"bytes")
        unpinned_name = faster_whisper_model_files("tiny")[0]
        unpinned_key = faster_whisper_manifest_key("tiny", unpinned_name)

        def fake_verify(path, filename=None):
            key = filename or os.path.basename(path)
            if key == unpinned_key:
                raise ChecksumError(f"No checksum is pinned for {key}")
            verify_model_file_real(path, filename)

        def fake_expected(filename):
            if os.path.basename(filename) == unpinned_key:
                return None
            return expected_for(filename)

        manager = _make_manager(engine="faster_whisper")
        manager.model_size = "tiny"
        manager._defer_download = True

        with patch(
            "vocalinux.speech_recognition.recognition_manager.faster_whisper.get_model_path",
            return_value=str(model_dir),
        ):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.verify_model_file",
                side_effect=fake_verify,
            ):
                with patch(
                    "vocalinux.speech_recognition.recognition_manager.expected_for",
                    side_effect=fake_expected,
                ):
                    with patch(
                        "vocalinux.speech_recognition.engines.faster_whisper_engine.FasterWhisperEngine"
                    ) as engine_cls:
                        manager._init_faster_whisper()

        assert (
            model_dir / unpinned_name
        ).exists(), "a missing pin is not a reason to delete the file"
        engine_cls.assert_not_called()
        assert manager._model_initialized is False


class TestFasterWhisperDownloadVerifiesExistingFiles:
    """Existence is not a pin: a leftover dest must still match its digest."""

    def test_an_existing_file_that_fails_its_pin_is_redownloaded(self, tmp_path):
        manager = _make_manager(engine="faster_whisper")
        manager.model_size = "tiny"
        model_dir = tmp_path / "tiny"
        model_dir.mkdir()

        first_name = faster_whisper_model_files("tiny")[0]
        first_file = model_dir / first_name
        first_file.write_bytes(b"not the file that is pinned")
        first_key = faster_whisper_manifest_key("tiny", first_name)

        streamed = []
        verify_calls = []

        def fake_stream(url, dest_path):
            streamed.append(dest_path)
            with open(dest_path, "wb") as handle:
                handle.write(b"good-enough")

        def fake_verify(path, filename=None):
            verify_calls.append((path, filename))
            if not str(path).endswith(".tmp"):
                raise ChecksumError("digest mismatch")

        mock_requests = MagicMock()
        mock_requests.exceptions.RequestException = Exception

        with patch.dict("sys.modules", {"requests": mock_requests}):
            with patch(
                "vocalinux.speech_recognition.recognition_manager.faster_whisper.get_model_path",
                return_value=str(model_dir),
            ):
                with patch.object(manager, "_stream_model_download", side_effect=fake_stream):
                    with patch(
                        "vocalinux.speech_recognition.recognition_manager.verify_model_file",
                        side_effect=fake_verify,
                    ):
                        manager._download_faster_whisper_model()

        assert any(
            path == str(first_file) and filename == first_key for path, filename in verify_calls
        ), "an existing dest must be hashed, not skipped because it is already on disk"
        assert any(
            os.path.basename(path) == first_name + ".tmp" for path in streamed
        ), "an existing bad file must be removed and re-downloaded"
        assert first_file.read_bytes() == b"good-enough"


class TestFailedReconfigureRestoresPreviousEngine:
    """A failed faster-whisper switch must not discard the live engine.

    Settings reverts the pickers to the saved config on error. If reconfigure
    keeps faster-whisper in ERROR after releasing whisper.cpp, dictation is
    dead until restart even though the UI shows the old engine.
    """

    def test_failed_faster_whisper_download_reloads_previous_engine(self):
        manager = _make_manager(engine="whisper_cpp")
        manager.engine = "whisper_cpp"
        manager.model_size = "tiny"
        manager._model_initialized = True
        manager.state = RecognitionState.IDLE

        with patch.object(
            manager, "_init_faster_whisper", side_effect=RuntimeError("download failed")
        ):
            with patch.object(manager, "_init_whispercpp") as restore_init:
                with pytest.raises(RuntimeError, match="download failed"):
                    manager.reconfigure(
                        engine="faster_whisper",
                        model_size="tiny",
                        force_download=True,
                    )
                restore_init.assert_called_once()

        assert manager.engine == "whisper_cpp"
        assert manager.model_size == "tiny"
        assert manager.state != RecognitionState.ERROR

    def test_restore_failure_stays_in_error(self):
        manager = _make_manager(engine="whisper_cpp")
        manager.engine = "whisper_cpp"
        manager.model_size = "tiny"
        manager.state = RecognitionState.IDLE

        with patch.object(
            manager, "_init_faster_whisper", side_effect=RuntimeError("download failed")
        ):
            with patch.object(
                manager, "_init_whispercpp", side_effect=RuntimeError("restore failed")
            ):
                with pytest.raises(RuntimeError, match="download failed"):
                    manager.reconfigure(
                        engine="faster_whisper",
                        model_size="tiny",
                        force_download=True,
                    )

        assert manager.engine == "whisper_cpp"
        assert manager.state == RecognitionState.ERROR

    def test_failed_switch_restores_live_settings_not_just_engine(self):
        """Unsaved VAD/device/API fields must roll back with the previous engine."""
        manager = _make_manager(engine="whisper_cpp")
        manager.engine = "whisper_cpp"
        manager.model_size = "tiny"
        manager.language = "en-us"
        manager.vad_sensitivity = 2
        manager.silence_timeout = 1.5
        manager.audio_device_index = 1
        manager.audio_device_name = "Built-in Mic"
        manager._voice_commands_preference = False
        manager.stop_sound_guard_ms = 200
        manager.whispercpp_n_threads = 4
        manager.whispercpp_gpu_device = 0
        manager.whispercpp_no_timestamps = True
        manager.remote_api_url = "http://old"
        manager.remote_api_key = "old-key"
        manager.remote_api_endpoint = "/inference"
        manager.remote_api_model = "whisper-1"
        manager._model_initialized = True
        manager.state = RecognitionState.IDLE

        with patch.object(
            manager, "_init_faster_whisper", side_effect=RuntimeError("download failed")
        ):
            with patch.object(manager, "_init_whispercpp") as restore_init:
                with pytest.raises(RuntimeError, match="download failed"):
                    manager.reconfigure(
                        engine="faster_whisper",
                        model_size="tiny",
                        vad_sensitivity=5,
                        silence_timeout=4.0,
                        audio_device_index=9,
                        audio_device_name="USB Mic",
                        voice_commands_enabled=True,
                        stop_sound_guard_ms=50,
                        whispercpp_n_threads=1,
                        whispercpp_gpu_device=2,
                        whispercpp_no_timestamps=False,
                        remote_api_url="http://new",
                        remote_api_key="new-key",
                        remote_api_endpoint="/v1/audio",
                        remote_api_model="sensevoice",
                        force_download=True,
                    )
                restore_init.assert_called_once()

        assert manager.engine == "whisper_cpp"
        assert manager.model_size == "tiny"
        assert manager.language == "en-us"
        assert manager.vad_sensitivity == 2
        assert manager.silence_timeout == 1.5
        assert manager.audio_device_index == 1
        assert manager.audio_device_name == "Built-in Mic"
        assert manager._voice_commands_preference is False
        assert manager._voice_commands_enabled is False
        assert manager.stop_sound_guard_ms == 200
        assert manager.whispercpp_n_threads == 4
        assert manager.whispercpp_gpu_device == 0
        assert manager.whispercpp_no_timestamps is True
        assert manager.remote_api_url == "http://old"
        assert manager.remote_api_key == "old-key"
        assert manager.remote_api_endpoint == "/inference"
        assert manager.remote_api_model == "whisper-1"
        assert manager.state != RecognitionState.ERROR


class TestAudioReconnection:
    """Test audio reconnection logic."""

    def test_attempt_audio_reconnection_success(self):
        """Test successful audio reconnection."""
        manager = _make_manager(engine="whisper_cpp")

        mock_pyaudio_mod = MagicMock()
        mock_pyaudio_mod.paInt16 = 8
        mock_stream = MagicMock()
        mock_stream.read.return_value = b"\x00" * 1024
        mock_audio_instance = MagicMock()
        mock_audio_instance.open.return_value = mock_stream

        with patch.dict("sys.modules", {"pyaudio": mock_pyaudio_mod}):
            with patch("time.sleep"):
                result = manager._attempt_audio_reconnection(mock_audio_instance)

        assert result is True
        assert manager._audio_stream == mock_stream

    def test_attempt_audio_reconnection_falls_back_to_default_resolver(self):
        """Test reconnection falls back when saved device name/index cannot resolve."""
        manager = _make_manager(engine="whisper_cpp", audio_device_name="Missing Mic")

        mock_pyaudio_mod = MagicMock()
        mock_pyaudio_mod.paInt16 = 8
        mock_stream = MagicMock()
        mock_stream.read.return_value = b"\x00" * 1024
        mock_audio_instance = MagicMock()
        mock_audio_instance.get_default_input_device_info.return_value = {"index": 0}
        mock_audio_instance.open.return_value = mock_stream

        with (
            patch.dict("sys.modules", {"pyaudio": mock_pyaudio_mod}),
            patch("time.sleep"),
            patch(
                "vocalinux.speech_recognition.recognition_manager._resolve_device_by_name",
                return_value=None,
            ) as mock_resolve_name,
            patch(
                "vocalinux.speech_recognition.recognition_manager._resolve_valid_input_device",
                return_value=1,
            ) as mock_resolve_default,
            patch(
                "vocalinux.speech_recognition.recognition_manager._get_supported_channels",
                return_value=1,
            ),
            patch(
                "vocalinux.speech_recognition.recognition_manager._get_supported_sample_rate",
                return_value=16000,
            ),
        ):
            result = manager._attempt_audio_reconnection(mock_audio_instance)

        assert result is True
        assert manager._audio_stream == mock_stream
        mock_resolve_name.assert_called_once_with(mock_audio_instance, "Missing Mic", None)
        mock_resolve_default.assert_called_once_with(mock_audio_instance, None)

    def test_attempt_audio_reconnection_no_resolved_device(self):
        """When no safe device is enumerated, reconnect via system default."""
        manager = _make_manager(engine="whisper_cpp", audio_device_name="Missing Mic")

        mock_pyaudio_mod = MagicMock()
        mock_pyaudio_mod.paInt16 = 8
        mock_audio_instance = MagicMock()
        mock_stream = MagicMock()

        with (
            patch.dict("sys.modules", {"pyaudio": mock_pyaudio_mod}),
            patch("time.sleep"),
            patch(
                "vocalinux.speech_recognition.recognition_manager._resolve_device_by_name",
                return_value=None,
            ),
            patch(
                "vocalinux.speech_recognition.recognition_manager._resolve_valid_input_device",
                return_value=None,
            ),
            patch(
                "vocalinux.speech_recognition.recognition_manager._open_capture_stream",
                return_value=(1, 16000, mock_stream),
            ) as mock_open,
        ):
            result = manager._attempt_audio_reconnection(mock_audio_instance)

        assert result is True
        mock_open.assert_called_once_with(mock_audio_instance, None)

    def test_attempt_audio_reconnection_max_attempts(self):
        """Test reconnection stops after max attempts."""
        manager = _make_manager(engine="whisper_cpp")
        manager._reconnection_attempts = manager._max_reconnection_attempts

        mock_audio_instance = MagicMock()

        with patch.dict("sys.modules", {"pyaudio": MagicMock()}):
            result = manager._attempt_audio_reconnection(mock_audio_instance)

        assert result is False

    def test_attempt_audio_reconnection_open_failure(self):
        """Test reconnection when stream open fails."""
        manager = _make_manager(engine="whisper_cpp")

        mock_pyaudio_mod = MagicMock()
        mock_pyaudio_mod.paInt16 = 8
        mock_audio_instance = MagicMock()
        mock_audio_instance.open.side_effect = IOError("Cannot open stream")

        with patch.dict("sys.modules", {"pyaudio": mock_pyaudio_mod}):
            with patch("time.sleep"):
                result = manager._attempt_audio_reconnection(mock_audio_instance)

        assert result is False

    def test_attempt_audio_reconnection_exponential_backoff(self):
        """Test exponential backoff in reconnection attempts."""
        manager = _make_manager(engine="whisper_cpp")
        manager._reconnection_delay = 0.1

        mock_pyaudio_mod = MagicMock()
        mock_pyaudio_mod.paInt16 = 8
        mock_stream = MagicMock()
        mock_stream.read.return_value = b"\x00" * 1024
        mock_audio_instance = MagicMock()
        mock_audio_instance.open.return_value = mock_stream

        sleep_durations = []

        def track_sleep(duration):
            sleep_durations.append(duration)

        with patch.dict("sys.modules", {"pyaudio": mock_pyaudio_mod}):
            with patch("time.sleep", side_effect=track_sleep):
                manager._reconnection_attempts = 0
                manager._attempt_audio_reconnection(mock_audio_instance)
                first_delay = sleep_durations[-1]

                manager._reconnection_attempts = 1
                manager._attempt_audio_reconnection(mock_audio_instance)
                second_delay = sleep_durations[-1]

        assert second_delay > first_delay
        assert second_delay == first_delay * 2

    def test_attempt_audio_reconnection_negotiation_fallback(self):
        """When negotiation returns no stream, reconnect falls back to plain open."""
        manager = _make_manager(engine="whisper_cpp")

        mock_pyaudio_mod = MagicMock()
        mock_pyaudio_mod.paInt16 = 8
        mock_stream = MagicMock()
        mock_stream.read.return_value = b"\x00" * 1024
        mock_audio_instance = MagicMock()
        mock_audio_instance.open.return_value = mock_stream

        with (
            patch.dict("sys.modules", {"pyaudio": mock_pyaudio_mod}),
            patch("time.sleep"),
            patch(
                "vocalinux.speech_recognition.recognition_manager._open_capture_stream",
                return_value=(1, 16000, None),
            ),
        ):
            result = manager._attempt_audio_reconnection(mock_audio_instance)

        assert result is True
        assert manager._audio_stream == mock_stream
        mock_audio_instance.open.assert_called_once()

    def test_attempt_audio_reconnection_empty_read_closes_stream(self):
        """A reconnected stream that returns no data must be closed safely."""
        manager = _make_manager(engine="whisper_cpp")

        mock_pyaudio_mod = MagicMock()
        mock_pyaudio_mod.paInt16 = 8
        mock_stream = MagicMock()
        mock_stream.read.return_value = b""
        mock_audio_instance = MagicMock()
        mock_audio_instance.open.return_value = mock_stream

        with patch.dict("sys.modules", {"pyaudio": mock_pyaudio_mod}):
            with patch("time.sleep"):
                result = manager._attempt_audio_reconnection(mock_audio_instance)

        assert result is False
        mock_stream.stop_stream.assert_called_once()
        mock_stream.close.assert_called_once()


class TestIBusEngineUtilities:
    """Test ibus_engine utility functions."""

    def test_is_ibus_available(self):
        """Test is_ibus_available() function."""
        from vocalinux.text_injection.ibus_engine import is_ibus_available

        result = is_ibus_available()
        assert isinstance(result, bool)

    def test_is_ibus_daemon_running(self):
        """Test daemon detection when not running."""
        from vocalinux.text_injection.ibus_engine import is_ibus_daemon_running

        with patch("subprocess.run") as mock_run:
            mock_run.return_value.returncode = 1
            result = is_ibus_daemon_running()
            assert result is False

    def test_is_ibus_daemon_running_success(self):
        """Test daemon detection when running."""
        from vocalinux.text_injection.ibus_engine import is_ibus_daemon_running

        with patch("subprocess.run") as mock_run:
            mock_run.return_value.returncode = 0